OpenWand Brings Context-Aware AI Collaboration to the Desktop

Summary

OpenWand is a free, cross-platform, MIT-licensed desktop application designed to keep AI alongside a user’s existing work. A global hotkey opens an overlay that can gather selected text, browser or application context, clipboard data, or a screen region, then send that material with a reusable or custom prompt to a configured cloud provider, local model, or OpenAI-compatible server. The application can stream answers into the overlay, render responses as local HTML and CSS, and support rewrite-in-place workflows that let users review proposed text before pasting it back into the active field. It also supports voice prompting and text-to-speech, screen snips for vision models, local optional memory, privacy warnings and redaction, and prompt-injection checks before requests are sent. OpenWand integrates with Codex and Claude Agent through their installed accounts and provides controls for permissions, project scope, continuity, progress, and conversation history. Its Python-based addon system can add prompts, hotkeys, context hooks, model tools, notifications, and other workflows in separate processes, while an MCP bridge exposes external MCP servers to the configured model. The project also includes a local read-only MCP context server for trusted clients such as Claude Desktop, Cursor, and Codex. Settings, chats, memory, and configuration remain on the user’s machine, while credentials are stored in the operating system keychain; model requests still go to the provider selected by the user. OpenWand runs on Windows 10+, macOS 13+, and Linux X11, with Wayland support still in progress. It requires Python 3.12 when run from source, and local models may need substantially more resources than the stated 4 GB minimum.